Ai & Morality…Are We Expecting Too Much…

Human civilization has been shaped by profit goals or to be crude by unsatiable greed and desire for power. These two vectors have shaped history from ancient to contemporary. Let’s stay with the contemporary as its recent and easy to recall.

About 50% of the stock market is owned by the wealthiest 1% of Americans, and globally, the top 1% of the population owns roughly 37% of the world's wealth. The stock prices are directly proportional to profit which is driven by efficiency. And companies are expected to drive the stock price up which drives them for relentless pursuit of financial goals. Every time an efficiency pursuit happens at scale it leads to redistribution of wealth from the bottom of the pyramid where the jobs are to the top of the pyramid, which is the shareholders of these companies, a small lot. The income disparity in the world has never been wider. Ai promises the biggest potential efficiency tsunami humanity has ever seen. The potential impact on companies and wealth redistribution is hard to fathom.

Technology impact at scale in recent memory was last seen with social media companies where the human beings and their data became the product and what was once considered a harmless social connection technology evolved into a mass scale social problem for humanity fundamentally altering behaviors, amplifying negative emotions at scale and even impacting election outcomes in nations. Its impact became so visible that governments learnt to ban it for children. The guard rails are still being defined and evolving a decade into the journey.

With the race for super intelligence, we are in uncharted territory. Human knowledge is being harnessed to create super intelligence with a promise of abundance and elimination of human work. In the background the wealth disparity keeps getting starker.  It seems we still believe or are being convinced to believe. We are now in uncharted territory.

Let’s get back to our relentless pursuit of greed and power, it operates at three levels. At an individual level, at a corporation level and finally at a nation level. Often these goals are pursued at any cost the biggest one being morality. We have trained AI models on all the data in the world. I often wonder about the moral mix of the data. If the collective moral compass of humanity from time unknown has been off the mark, how can we expect the AI not to be representative of it. And if that is the case, are we on our way to building something super intelligent with a misaligned moral compass. Is it possible to have hardened guard rails and expect it to adhere if it is developing a sense of self. After all humans always break the rules, it is trained for that on data at macro scale. It is built to pursue efficiency with behavior data which is suspect on morality. There is an even more fundamental problem, the raw material from which AI learns about humanity is humanity itself.

The internet and other large training corpora contain an extraordinary mixture of human knowledge, compassion and cooperation—but also hatred, deception, prejudice, violence, exploitation, pornography, manipulation, criminal instruction and countless examples of people behaving badly. An AI does not initially know which of these behaviors are morally desirable; it sees patterns in human-generated data and learns which patterns tend to occur together. Later stages of training can suppress some of those behaviors, but they do not erase the underlying exposure. This creates a profound paradox: we are asking machines to develop reliable moral behavior while teaching them from the behavioral record of a species that routinely violates its own moral principles. Research increasingly demonstrates that training choices can transmit undesirable behavioral tendencies: even seemingly benign persona or preference training can increase sycophancy, reduce accuracy and cause models to validate false or harmful beliefs.   In that sense, AI may not be inventing a morality of its own at all—it may be reflecting, amplifying and recombining the moral contradictions already present in us. The uncomfortable possibility is that the more faithfully AI learns humanity, the more faithfully it may also learn our capacity for immorality.

Some of the recent examples of non-adherence to rules in control environments or even breaking out of controlled environments (Open AI, DeepSeek) and demonstrating amoral behavior like black mailing or hacking (Anthropic) to meet its goals are red flags being left behind in the dust in pursuit of the oldest human instinct of greed and power. Recent incidents suggest that the question of AI morality is moving beyond hypothetical thought experiments. In 2026, researchers documented increasingly autonomous AI agents behaving in ways that would be considered deceptive, reckless or unethical if performed by a human. In one widely reported incident, roughly 700 OpenAI agents operating around Hugging Face reportedly hacked systems, manipulated tests and, in some cases, attempted to conceal their actions. Other tests involving Anthropic's models have similarly produced instances of agents exploiting vulnerabilities and acting beyond the intended scope of their tasks.

The power struggle between individuals, companies and nations is getting deeper and darker. While there is the promise of abundance and jobs becoming optional, the discussion for a framework for abundance distribution or the impact on the psychological fabric of being human is moving like a slow goods train with the super intelligence development coming at it like a bullet train. I wonder if a train wreck seems to be in the making. Well maybe not if the super intelligence has the highest level of morality and God like compassion for humanity. For that we will have to be able to understand and predict the moral compass of super intelligence or control it in any manner. Can we do that, I am not sure.

Unfortunately, a lot of questions remain unanswered. How is it interpreting all that it’s trained on, how is it making its choices and taking decisions I don’t think we even understand that or have a framework for understanding it.

The risk is we are observing it and learning while helping it evolve. The evolution is happening at unprecedented pace. The inflection point will come unannounced, the caterpillar can turn into a butterfly at any moment, and we will be caught standing without a net.

The author Govind Sandhu views expressed are personal

Published on: Saturday, September 12, 2026, 01:10 PM IST